Speaker Recognition System Based on GMM using LPOcepstrum and Pitch

摘要

This paper presents the experimental comparison of a speaker recognition system based on Gaussian Mixtures Model (GMM) by reinforcing and enhancing the feature vector characteristics. The basehne system proposed use a GMM with 8 gaussian mixtures with diagonal covariance matrix and for feature vector the LPC-cepstrum coefficients are used. One system is developed to improve the recognition rate by using only voiced part to extract the LPCxepstrum coefficients. Other system is made reinforcing the feature vector LPC-cepstrum with pitch information. The system is compared with the system enhanced by the feature vector LPC-cepstrum with Cepstral Mean Normalization (CMN). The experimental comparisons of those systems are evaluated in closed and open test using speech data with time difference.